What drives rental levels in the non-regulated rental housing market of Amsterdam? Do there exist differences between submarkets or between Amsterdam and
Rotterdam? A hedonic regression model.
Student: Rutger W. de Vries
Student number: 10003400
Thesis supervisor: dhr. dr. M.A.J. Theebe
Faculty: Faculty Economics and Business
Study: MSc Business Economics
Acknowledgements
I would like to take this opportunity to thank my family and my lovely girlfriend for their great support during the last months of my study at the University of
Amsterdam. Special thanks go to my parents who have given me the opportunity to study. In addition, I wish to express my sincere thanks to my supervisor dr. M.A.J. Theebe for his clear feedback, contribution and patience during the process of writing this master thesis. At last, I would like to thank Pararius for providing me extensive data of the Dutch housing market.
Statement of Originality
This document is written by Student Rutger de Vries who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Table of contents
1.#Introduction#...#5! 1.1! Increasing#interest#in#the#Dutch#rented#housing#sector#...#5! 1.2! Research#method#...#7! 1.3! Research#questions#...#8! 1.4! Thesis#content#...#10! 2.#Review#Dutch#residential#housing#market#...#11! 2.1#Dutch#housing#market#...#11! 2.2#Amsterdam#...#13! 2.3#Rotterdam#...#14! 3.#Related#literature#...#16! 3.1#Introduction#related#literature#...#16! 3.2#Hedonic#pricing#model#...#17! 3.3#Rent#level#literature#...#17! 3.4#Conclusion#related#literature#...#27! 4.#Research#methodology#...#30! 4.1#Hedonic#regression#model#...#30! 4.1.1.!Rent!level!drivers!in!Amsterdam!...!32! 4.1.2.!Rent!levels!drivers!in!submarkets!in!Amsterdam!...!32! 4.1.3.!A!comparison!between!Amsterdam!&!Rotterdam!...!33! 4.1.4.!A!comparison!between!submarkets!in!Amsterdam!&!Rotterdam!...!34! 4.2#Variables#...#35! 4.3#Assumptions#...#39! 4.4#Considerations#...#39! 4.4.1!Multicollinearity!...!39! 4.4.2!Heteroskedasticity!...!40! 5.#Data#...#41! 5.1#Data#preparation#...#41! 5.2#Descriptive#statistics#...#42! 5.2.1!House!types!...!44! 5.2.2!City!districts!...!45! 5.2.3!Rent!segments!...!47! 5.2.4!Rents!over!time!...!48! 5.2.5!MeanFcomparison!tests!submarkets!...!49! 5.3#Conclusion#summary#statistics#...#51! 6. Results#...#52! 6.1#Multicollinearity#&#heteroskedasticity#...#52!6.2.1!Rent!level!drivers!in!Amsterdam!...!53! 6.2.2!Rent!level!drivers!in!submarkets!in!Amsterdam!...!55! 6.2.3!A!comparison!between!Amsterdam!&!Rotterdam!...!63! 6.2.4!A!comparison!between!submarkets!in!Amsterdam!&!Rotterdam!...!64! 6.4#Regression#with#interaction#terms#...#67! 6.5#Additional#regressions#...#75! 7.#Conclusion#...#77! Further#research#...#79! References#...#80! Appendix#...#82!
1.#Introduction!!!
1.1 Increasing interest in the Dutch rented housing sector !
The interest among investors in rental housing has skyrocketed in the Netherlands during the last years. This is supported by a recently published report of real estate advisory firm Capital Value (2014). Based on this research Dutch investors have approximately €2 billion available for investments in rented housing in the Netherlands, while foreign investors want to invest approximately €1 billion in 2014. Among 200 international real estate investors from North America, Asia, the Middle East and Europe, 29% are actively pursuing acquisitions in the Netherlands (Capital Value, 2014). The interest among national and international investors in the Dutch residential housing markets increases the importance of both fair asking rents and insight in the structure of this fair asking rent levels.
BNP Paribas REIM Germany for instance already acquired a portfolio of residential properties in the Randstad from Amvest with a value of €40 million in 2014 (PropertyNL). In addition, German investment firm Patrizia bought 5.500 rented residential properties from housing association Vestia for €578 million in mid 2014 as well (FD, 2014). Finally, Round Hill Capital acquired almost 3,800 residential properties from the Dutch ‘Wooninvesteringsfonds’ (WIF) for a value of €365 million in the end of 2014. In June 2014, Round Hill Capital already acquired approximately 1,500 dwellings for €180 million from CBRE GI (PropertyNL).
Previously, commercial investors were not highly active in the Dutch rental housing market since 80% of rental housing was controlled by Dutch housing associations and direct returns were low (Ministery of Foreign Affairs, 2013). However, housing associations are currently selling large blocks of regulated and non-regulated residential properties to the market (FD, 29-10-2014). Vestia for example planned to sell 30.000 dwellings before 2020. Based on CBRE (2014), a substantial part of social housing is eligible for the non-regulated sector. According to PropertyNL (2014), this part sums up to almost 50% of total social housing. This
transformation is in line with the objective of the Dutch government to increase the non-regulated housing sector nowadays.
These market developments provide interesting purchasing opportunities to residential real estate investors. Based on research of CBRE in 2014, increasing interest in the Dutch housing market is a result of improved market conditions, bottomed-out house prices, attractive risk-return ratio, stable income returns and an increasing shortage on the residential real estate market. Interestingly, demand for non-regulated rental housing will increase by 200.000 until 2020 based on a report of Deloitte (2012).
Especially rental apartment complexes in the Netherlands are popular for investors due to low management costs, low vacancy risk and positive demographic and governmental developments (Capital Value, 2014). Institutional investors are currently heavily focusing on the acquisition of rental dwellings in urban growth regions, which are non-regulated or can be deregulated after the purchase, to avoid any landlord levy.
Rents of non-regulated rental housing are, in contrast to regulated rental housing, not controlled and regulated by the Dutch government and must reflect market rents. Because of the increasing opportunities and interest among commercial investors in the Dutch housing market, it is important to use an appropriate valuation method for residential properties. Therefore, this thesis will provide an empirical valuation model that could be used as a suitable tool to determine fair asking rents and observe important rent level drivers of non-regulated rental dwellings in the Netherlands.
Especially, the housing market of Amsterdam is strongly in demand among investors. The housing market in Amsterdam has to deal with a substantial housing shortage. Besides this, Amsterdam is the capital city and the Amsterdam Metropolitan Area (MRA) is listed on the 5th place ranked on economic importance in Europe (MRA, 2014). The MRA will therefore face strong economic and employment growth. In the first quarter of 2014, 15% of total residential real estate investments in the Netherlands was made only in Amsterdam (CBRE, 2014). CBRE expects this share
will increase thanks to the sale of recently constructed developments in Amsterdam. These favorable developments in Amsterdam provide attractive investment opportunities to investors nowadays. Because of the large investment share of Amsterdam and proper data availability, the rental housing market of Amsterdam is examined to observe rental drivers and determine fair asking rents of rental units. Since the housing market of Rotterdam has the largest and equal share of non-regulated rental housing, this market is used as a control group for Amsterdam in this thesis.
1.2 Research method !
Real estate has to deal with substantial quality differences between properties as a result of age and size for instance. Therefore it is difficult to compare rent levels of dwellings in two or more different areas (Hoesli, Thion and Watkins, 1997). To correct for heterogeneity among rental dwellings, a hedonic regression model is applied in this thesis. The hedonic regression model focuses on heterogeneous non-regulated rental dwellings in Amsterdam and examines the specific impact of rent level drivers on these rents. The impact of house specific characteristics, house features, house types, locational variables, listing period and time effects on rental levels in Amsterdam is analyzed. Based on Allen, Springer and Waller (1995): ‘the periodic rent for a rental property reflects the market price that users are willing to pay for the rights to the package of real estate services for a specified amount of time’. The empirical model is generally employed to construct house price indexes and examine the relationship between house prices and house price drivers. This thesis however will show that the model could also be applied to the rental housing sector. The model must provide understanding of the structure of non-regulated rental housing markets. Concluding, this thesis provides real estate investors a suitable model that analyses rent level drivers and determines the fair asking rent of rental dwellings in Amsterdam based on their underlying attributes.
Firstly, according to existing real estate literature rent levels play a crucial role in the field of property valuation. Rent levels are a fundamental determinant of the
investment value of a property (Gallin, 2008). Fair market rents are therefore
necessary to apply the income valuation method on properties. The hedonic model in this thesis could therefore be a suitable tool to provide the fair value of the property based on the fair asking rents of these rental dwellings.
Secondly, the hedonic model could be applied by residential real estate investors to support their investment policy. With the aid of the hedonic model, investors are able to seek for residential real estate with potentially the highest initial asking rents and therefore the highest gross initial yield. Therefore, the hedonic model could determine the direct return on the investment for residential real estate
investors.
Thirdly, the examination of rent level drivers provides residential property investorsa clear insight and understanding of factors that significantly affect residential property rents. Studying the impact of rent level drivers could also determine the profitability of providing additional services or amenities to a tenant.
In addition, Guntermann and Norrbin (1987) noted that the application of such a hedonic regression model could be used by appraisers for rent adjustments, by property managers for determining rents and by feasibility analysts for designing new apartment projects. Finally, based on Hoesli et al. (1997), the hedonic regression model is easy to use.
1.3 Research questions
With!the!aid!of!the!discussed!research!method!the!following!research!questions! will!be!answered!in!this!thesis.!The!main!research!question!is!listed!below:!
(1) What are the most important rent level drivers in the non-regulated housing sector in Amsterdam?
This thesis also examines if there exists market segmentation in the non-regulated rental housing sector as in line with previous literature To capture submarket-specific differences, rent level drivers are tested separately for different house types and rent segments. It is likely that rent levels react different to house specific characteristics in various rent segments. In addition, several rent level drivers could be more valuable
for some house type. Based on Guntermann and Norrbin (1987), access to open space such as balcony, garden or roofterrace might be more valuable for apartments because of the density. The following sub question is formulated to research if rent level drivers differ per submarket.
(2) Does the impact of rent level drivers differ between house types or rent segments in Amsterdam?
In recent years Amsterdam has faced stronger rent increases compared to other major cities in the Netherlands like; Utrecht, Den Haag and Rotterdam (Pararius, 2014). Based on a recently published report by Pararius, the average rent in Amsterdam was €19 per square meter in the second quarter of 2014 while the national average rent level in the Netherlands was €12 per square meter. To analyze the substantial rent difference between Amsterdam and other major cities in the Netherlands, this thesis will compare the impact of underlying rent level drivers between the two largest cities in the Netherlands, Amsterdam and Rotterdam. The additional aim therefore is to examine if there exist regional differences in the Dutch rental housing market based on underlying rent level drivers. Amsterdam and Rotterdam consist of 430.000 and 310.000 households respectively (CBS, 2014). Furthermore, both cities have the largest non-regulated rental housing sector and the best data availability in the Netherlands. The average rent level in Rotterdam is equal to the national average consisting of €12 per square meter.
Firstly, rent level drivers of rental units in Rotterdam are researched with the aid of the base hedonic regression model. In addition, rent level drivers in Rotterdam are tested separately for different submarkets as explained earlier in this section. To correctly measure whether rent level drivers between Amsterdam and Rotterdam differ significantly during the sample period, a Chow-test is performed. Besides this, a regression model including interaction terms is created to specifically measure the differences in effect of rent level drivers between Amsterdam and Rotterdam. The additional sub questions are formulated as follows:
(3) Does the impact of rent level drivers differ between Amsterdam and Rotterdam?
(4)Does the impact of rent level drivers differ between house types or rent segments in Amsterdam and Rotterdam?
The main contribution of this thesis is that existing literature about non-regulated rental housing is scarce in the Netherlands. This is mainly due to the unavailability of rental levels and underlying house attributes. ‘Market rents cannot easily be derived from a hedonic price model, because the non-regulated rental market is very small, so market rents are hardly available’ (Francke, 2010). Private rental levels are not centrally registered and therefore not publicly available. This thesis however has access to asking rents of more than 20.000 rental dwellings in both Amsterdam and Rotterdam for the period Q2 2011 – Q2 2014. In addition, to existing literature this thesis focuses on both apartments and single-family houses. A hedonic regression analysis of cross-sectional non-regulated rental units could therefore be a relevant contribution to existing residential real estate literature.
1.4 Thesis content !
Firstly, section 2 will shortly describe the current situation on the Dutch housing market. Section 3 will analyze existing related literature. Subsequently, section 4 will further elaborate on the research methodology. Section 5 will focus on data and descriptive statistics. In section 6 regression results are showed. Finally, section 7 will provide a conclusion.
2.#Review#Dutch&residential&housing(
market!!
This section will shortly describe the current situation on the Dutch residential housing market to improve the understanding of this market. Firstly, the basic characteristics of the housing market are discussed. Secondly, the current situation in Amsterdam and Rotterdam is briefly illustrated.
2.1 Dutch housing market !
The Dutch housing market is characterized by intensive governmental regulation, subsidization, a relatively low home-ownership rate and detailed spatial planning. Governmental policies had a strong focus to increase homeownership in the owner-occupied housing market, which resulted in an increase from 36% to 60% since 1972. Because of land regulation new construction is restricted in suburban areas. Therefore the Dutch housing market has a remarkably low supply elasticity of housing (Vandevyvere & Zenthöfer, 2012).
The Dutch government also provides social housing to low-income groups and regulates through rent control in the rental housing market. The large supply of social housing in the Netherlands is a result of the Housing act (1901), which provides housing and improves housing conditions of Dutch households. In addition, the Dutch government provides housing allowances to low-income households which pay a monthly rent up to €699 (Conijn, 2014). Rental housing in the Netherlands with a rent above this liberalization rent limit are part of the non-regulated housing sector. This is based on a housing evaluation system (WWS) that assigns points based on housing and neighborhood characteristics and determines the maximum reasonable rent. The share of the non-regulated rental housing sector in the Netherlands is one of the smallest in the world (FD, 30-10-2014). As mentioned before, real estate investors were therefore not highly active in the Dutch housing market.
On the one side, there exist large differences between rental levels in the regulated- and non-regulated sector because of provided social housing by Dutch
housing associations. Rental levels in the social housing sector do not reflect market conditions due to rent regulation and rent control (Francke, 2010). Pararius, the largest online supplier of rental housing published a report at the end of 2013 stating that the average monthly rent in the private sector of Amsterdam was €1.824. In contrast to private rents, the average monthly rent in Amsterdam in the social housing sector was approximately €500 during 2013, which equals national average.
On the other side, there exist a substantial value gap in the Dutch housing market according to existing housing literature of Conijn (2014). This means house prices are too high due to the capitalized subsidies, while the majority of rental levels is far too low due to extensive social housing, rent regulation and housing allowances. The Dutch government namely implicitly subsidizes the owner-occupied housing market as well through mortgage interest deductibility to stimulate homeownership. 2014). Although, since the financial crisis in the Netherlands house prices have declined by approximately 20% and rents have increased by 5% (Capital Value, 2014).
However, the Dutch government currently progressively encourages the increase of the private rental housing sector in the Netherlands. The government gradually limits interest tax deductibility for owner-occupiers, raises rents for high income households in social housing and stimulates additional non-regulated rental housing construction and the transformation of vacant office space into non-regulated rental housing (FD, 29-09-2014).
Based on BouwInvest (2013) the total housing market in the Netherlands consists of approximately 7.2 million houses of which 60% is owner-occupied and 40% is rented. The total share of social housing in the Dutch housing market consists of almost 33% of total housing stock. Based on a report of Deloitte (2012) the non-regulated housing market consists of approximately 3% of total housing supply in the Netherlands. Therefore, the share of the private rental housing market is one of the smallest in Europe, where France and Germany have a share of 24% and 54% respectively (Deloitte, 2012). In addition, the Netherlands has the highest share of social housing in Europe (Vandevyvere & Zenthöfer, 2012). Figure 1 provides insight in the current Dutch housing market. Only 2% of rental housing owned by housing associations is
non-regulated, while 30% of rental housing owned by investors is non-regulated. However, the share of institutional investors in the housing market is relatively high compared to other European countries. This implies a more liquid housing market, which is attractive for investors (FD, 22-10-2014).
Figure 1: The Dutch housing market
(Capital Value, 2014)
2.2 Amsterdam !
The housing market in Amsterdam is characterized by a substantial housing shortage, even in comparison to the housing market of Rotterdam. This is mainly caused by rapidly increasing inhabitants and single households (CBS, 2014). The housing market in Amsterdam consists of much more rental dwellings in comparison with the national average. Figure 2 listed below explains the current division of owner-occupied housing, social housing and private rental housing in Amsterdam (CBRE, 2014). Almost 70% of the residential properties in Amsterdam are rental dwellings, whereof approximately 30% is not regulated by the Dutch government. Amsterdam has a population of 780.000 habitants and 430.000 households (United Nations, 2014).
Figure 2: The housing market of Amsterdam
(CBRE, 2014)
In line with other European countries, the Netherlands faced substantial house price drops since the outbreak of the financial crisis in 2008. The average house price in the Netherlands is currently €224,000. Amsterdam has a slightly higher house price, namely €238.000 (CBS, 2014).
2.3 Rotterdam !
Based on figures of COS Rotterdam (2013), the rental housing sector in Rotterdam consists of approximately 70% of total housing. This share is equal to the percentage of rental dwellings in Amsterdam. 20% of total rental housing is non-regulated, which is almost equal to Amsterdam. Figure 3 summarizes these findings. The division between owner-occupied housing, social housing and private rental housing provides a proper starting point for the comparison between Amsterdam and Rotterdam. The average house price in Rotterdam currently is €148.000, which is substantially lower than the house price in Amsterdam. Furthermore, Rotterdam has 610.000 inhabitants and 310.000 households (CBS, 2014). These figures imply less habitants on average per dwelling.
Figure 3: The housing market of Rotterdam
(COS Rotterdam, 2013)
Figure 4 illustrates the nominal rent level per square meter of five large cities in the Netherlands for the period Q1 2007 to Q3 2014. As mentioned in the introduction, a large gap exists between rent levels of Amsterdam and Rotterdam. Amsterdam has a rent level per m2 of approximately €19, where Rotterdam has a rent level per m2 of approximately €12. This provides an interesting starting point for the examination of rent level drivers in both cities. Section 5.2 explains differences between both cities in further detail.
Figure 4: Nominal monthly rent levels per square meter (2007-2014)
3."Related'literature!!
3.1 Introduction related literature !
The following figure shows various interconnected factors that effect residential housing markets in general. These drivers are short-term or long-term orientated. The economy and demography are drivers on the demand side of the market. On the other side, construction and real estate stock determine the supply side of the residential housing market. This section however mainly focuses on existing rental housing literature explaining non-regulated rental levels specifically based on their underlying attributes. In comparison with hedonic house price literature hedonic rent level literature is relatively scarce. Therefore, a few relevant findings from house price literature are shortly summarized as well. The following related literature is ranked chronologically.
Figure 5: Drivers residential housing market
The contribution of existing related literature is that important house specific rent level drivers could be analyzed. Table 1 listed in the Appendix summarizes the specific impact of significant explanatory variables in former hedonic research. This provides a starting point for constructing the hedonic model in this thesis.
3.2 Hedonic pricing model !
The hedonic pricing model was introduced by Court in 1939. Court (1939) examined the relationship between the enjoyment of an automobile and the underlying attributes of the automobile. The main aim of a hedonic regression model is to link the price of a good to its underlying characteristics. Each individual characteristic contributes to the overall value of the underlying object (Francke, 2013). Based on Goodman (1998), the term hedonic means: ‘the weighting of the relative importance of various components among others in constructing an index of usefulness and desirability’.
Rosen (1974) was the first researcher who applied the hedonic pricing theory to the residential housing market. This researcher showed that the hedonic regression model is applicable to effectively value apartments. Each specific house attribute determines value of the dwelling. Since house characteristics cannot be traded separately, Rosen (1974) examined the marginal contribution of these property specific characteristics. The application of the hedonic regression model by Rosen (1974) turned out to be the foundation for existing research in the field of real estate valuation. However, many researchers disagree about the structure of the functional form in the hedonic regression model.
The hedonic price function provides a suitable framework for the valuation of differentiated heterogeneous residential properties, whereby individual attributes do not have an observable market price. In real estate literature the hedonic pricing model is a form of an advanced valuation model (Selim, 2009).
3.3 Rent level literature !
The first paper that is discussed is the one of Marks (1984). Marks performed a hedonic regression model to explain the variation in rents of both controlled- and
level drivers, data of almost 4.000 apartments from 1978 are included in his paper. Outside the scope of this thesis, Marks (1984) studied the impact of governmental rent control on rent levels of regulated rental housing. This thesis however will focus on the hedonic regression explaining the variation in rents of uncontrolled rental units. In Vancouver only 20% of total rental housing is non-regulated by the Canadian government. According to Marks (1984), results could be biased because data on unregulated apartments is not monitored as precisely as data on controlled apartments and the sample is relatively small with 687 observations.
As listed in table 1, Marks (1984) included locational variables, structural building characteristics, additional features and the age of the apartment in the regression to explain variation in rents. In addition, Marks (1984) added a time dummy to the hedonic model to measure time effects. Both a linear and a semi-logarithmic functional form have been applied in his paper.
The most statistically significant variables are furniture, the number of bedrooms and a location dummy east/west, meaning that rents are substantially higher in the western part of Vancouver. Furniture increases rents by almost 50% ceteris paribus. Based on Marks (1984), this is probably due to recent renovation of furnished units in the uncontrolled housing sector of Vancouver. In addition, Marks (1984) found that the number of stories has a significantly positive impact on rents due to water- and mountain-view in Vancouver. An additional bedroom increases rents of uncontrolled units with 15%. The impact of the distance variable is significantly negative since apartments situated closer to the city center have higher rents. Marks also included the unique variable resident caretaker, which turned out to have a significantly positive impact on rents. At last, Marks incorporated the time dummy in the regression. This time dummy indicated a particular quarter of the year to control for seasonality or cyclical influences on rents (Marks, 1984). The variable Q1 turned out to be significantly negative. His model explained more than 60% of total variance.
Guntermann and Norbin (1987) recognized that explaining variation in apartment rents is an overlooked topic in literature. The researchers emphasize the importance of such empirical models, because the models could be useful for appraisers to estimate the value of specific amenities and for property managers to determine apartment
rents. Besides this, the model is a useful application for feasibility studies when analyzing various features and amenities for new projects. Guntermann and Norrbin (1987) define residential real estate rents as the direct market estimate of the value of housing services based on characteristics, amenities and locational attributes.
Guntermann and Norrbin (1987) studied rent variation of apartments in the Phoenix metropolitan area. In their research various physical characteristics, amenities and locational attributes are analyzed. However, they examined rent variation of rental units in different submarkets as well. To analyze rent level drivers separately, a university market is tested in comparison to a non-university market and a market with new projects is tested against a market with old-fashioned projects. At last, high-rent rental units are tested in comparison with low-rent rental units.
The dataset consists of market rents covering 291 rental units, which were gathered from resident managers in the Phoenix metropolitan area during the first quarter of 1984.
In contrast to other researchers Guntermann and Norrbin (1987) applied a log-linear specification to deal with heteroskedasticity in the cross-sectional data. They also addressed the important issue of multicollinearity caused by the large number of explanatory variables in the hedonic regression model. In their research multicollinearity results in sign reversals and insignificance of facility variables like sauna’s or swimming pools. Therefore, only a few powerful components are included in the regression. Table 1 listed rent level drivers mentioned in their paper.
The performed model of Gunterrman and Norrbin (1987) has good explanatory power, namely 81%. Especially the variable size is highly significant and powerful in explaining rents. The marginal impact of an additional bedroom is tested as well for studios, one- and two-bedroom apartments. The strongest increase they found was for an additional bedroom in a two-bedroom apartment, namely a 10% increase in rents.
The variable age has a small impact on rent levels. A possible explanation based on the paper of Gunterrman and Norrbin (1987) could be that the condition and the level of maintenance is much more relevant in explaining rents. Therefore, they also included the variable condition, which is a measurement determined by resident
The impact of the variable bathrooms is significantly positive with an effect of 2.7%. Guntermann and Norrbin created this variable to measure the impact of an additional bathroom beyond the standard one bath per two bedrooms. In addition, apartments with a pool or fireplace cost 7.8% and 7% more respectively. Finally, Guntermann and Norrbin (1987) included proximity proxies to measure accessibility to the
highway ramp or to downtown Phoenix. A striking result is that rents and distance to the center are positively related, probably caused by the construction of luxurious apartment complexes far from the city center.
Finally, regression results for different submarkets are explained. Guntermann and Norrbin (1987) analyzed on the one side a submarket consisting of rental units located within two miles from the University compared to rental units located further than two miles from the University. On the other side they studied a project age market with apartments older than four years in comparison with apartments less than four years old. In addition, high- and low-rent rental units were compared.
The results show that distance to the university has more impact on rents than the condition of the rental unit in a student-market. Based on existing house price literature, accessibility to the university generally has high explanatory power. A Chow-test illustrates that the impact of common area amenities is significantly higher for student housing within two miles from the university. Furthermore, their results also showed that there exist substantial quality differences between newer projects, in contrast to older projects, which results in higher rents. House characteristics such as fireplace or dishwasher were more powerful for older projects in explaining rental levels compared to newer projects. This is due to the fact that more recently built projects are likely to already have these amenities. Finally, no significant differences were found between high- and low-rent rental units.
Two years later, Sirmans, Sirmans and Benjamin (1989) published an article in the journal of real estate research introducing external effects. They studied the impact of several rent level drivers to determine apartment rents, which are listed in table 1. Sirmans et al. (1989) examined these drivers with an extensive hedonic regression model including physical-, locational-, amenity-, service- and external factors.
Sirmans et al. (1989) included expenses that are passed through to the tenant in rents. These researchers combined or transformed variables to cope with multicollinearity.
Their dataset consists of 188 observations from 92 residential apartment complexes in Louisiana, America. In this research both linear and semi-logarithmic functional forms are applied which explain 67% and 61% of total variance respectively. Sirmans et al. (1989) found that the size of a property is the most powerful driver of rents per rental unit as listed in table 1. Due to a lack of data on specific apartment size, property size is measured by the number of bedrooms. However in contrast to the paper of Marks (1984), Sirmans et al. (1989) measure the marginal effect of an additional bedroom. Especially, the impact of a second or a third bedroom is highly significant in explaining rents. The largest marginal rent increase for a third bedroom is observed in the semi-log model, where rents increase with 57%. This is due to the fact that a third bedroom implies additional bathrooms and a larger common area, such as a living room or a dining room (Sirmans et al., 1989). Furthermore, the existence of a covered parking place or a swimming pool strongly increases the rents with effects of 15% and 5% respectively. Results of the linear and semi-log functional form turned out to be quite similar, especially the signs of the coefficients.
As discussed, Sirmans et al. (1989) also included external factors in the hedonic model. These factors are the accessibility to public transport and congestion in the surroundings. The accessibility to public transport is expressed as a zero-one dummy variable, which measures if an apartment is located within thousand feet of a bus line. Congestion is measured as the distance from a major arterial street. Both variables are significant in explaining rents and have a negative sign.
The size of the apartment complex as a whole has a positive significant impact on rents per rental unit since tenants prefer social interaction. Larger complexes generally have more amenities and services as well. The inclusion of complex size is often applied in further research. The restriction of pets has a significantly negative effect on rents. This is caused by the fact that a premium has to be paid for the permission of keeping pets. Finally, Sirmans et al. (1989) included location dummies as well. Two specific areas in the city have a significantly positive impact on rental
In addition to the hedonic pricing model, Sirmans et al. (1989) included a cost/benefit analysis in their paper. The main aim of this methodical analysis is to determine the profitability of providing additional services or amenities to the tenant. Based on this cost/benefit analysis the addition of a covered parking place results in a profit for the landlord.
In accordance with Guntermann and Norrbin (1987), Allen, Springer and Waller (1995) also recognized market segmentation in the rental housing market. The submarkets are determined by property type and consist of apartments, condominiums and single-family houses. To avoid violating the OLS assumptions, these three submarkets have to be tested separately instead of an aggregate rental housing market as a whole. Therefore, they studied price differences of rental housing characteristics across the three aforementioned property types. For better comparability of the submarkets, they only include parameters into the empirical model in case this variable affects all three property types. Allen et al. (1995) found that the impact of various rent level drivers differs across property type. Therefore, this part of the related literature mainly focuses on the examination of different submarkets.
Their sample consists of approximately 1,300 rental units from Clemson (South Carolina) covering 1991. The following model is applied: !!" = ! !!, !!, !!, !!, !! , where (P) are property physical characteristics, (A) are available amenities, (L) are location characteristics, (T) are tenant characteristics and (M) are property management arrangements. As dependent variable in the condominium market, both stated rents and rents including condominium fees are employed. Allen et al. (1995) tested for rents and rents per square foot in natural log form. Rent level drivers again listed in table 1 in the Appendix.
Allen et al. (1995) found that square feet and distance were highly significant in explaining rents in all submarkets. In their research size is again measured by square feet. Distance measures the distance to the University, where the University serves as a proxy of economic center of the city. Based on the mono-centric model, rents and distance have an inverse relationship.
The most powerful explanatory variable is the number of bathrooms. The impact of an additional bathroom on rents is the strongest for apartments, with an
effect of 24%. The impact of an additional bathroom on single-family house rents is only 12%. Furthermore, the impact of additional square feet is 2% for apartments and 1.5% for single-family houses, while the effect on the whole sample is only 1.3%. Property age is negatively related to rents of apartments and single-family houses. This is in accordance with previous research. Finally, the impact of an additional bathroom or furniture on rental units measured with OLS in the whole sample is 16% and 20% respectively.
Allen et al. (1995) included several innovative variables in their empirical model. Firstly, they measured furniture of the rental dwelling. However, the impact of furniture is only significant for the condominium market. The researchers also include uncommon variables as tenant characteristics, tenant tenure and property
management arrangement. These variables are not often used in literature because of
scarce data availability. Allen et al. (1995) found that only for single-family houses students have to pay a rent premium. Hoesli et al. (1997) will study tenant characteristics once again. Furthermore, results show that tenant tenure is not significant in any submarket and a local owner receives higher rents in the condominium market. Finally, no large differences are found between the model with stated rents of condominiums and rents including condominium fees (Allen et al., 1995).
At last, Allen et al. (1995) showed that all three models differ significantly from one another at a 1% significance level with the aid of a Chow-test. In addition, a Tiao-Goldberger test illustrated that the variables square feet, furniture, student
tenants and local owner differ significantly across all submarkets.
Hoesli, Thion and Watkins (1997) studied the impact of only a few physical characteristics and environmental variables on rental levels as listed in table 1. These researchers also recognized market segmentation in the form of differences across neighborhoods. Their main aim was to show that the hedonic regression technique could also have been applied properly to the private rental housing sector of Bordeaux.
was gathered from real estate agencies in Bordeaux during 1994. In accordance to Sirmans et al. (1989), rents per unit include expenses that are passed through to the tenant. This adjustment improves the comparability across rental units.
Hoesli et al. (1997) performed a hedonic Pricing Model on the basis of both a linear specification and a logarithmic form. Empirical results turned out to be quite similar. In their paper the issue of multicollinearity is mentioned. Property size and the number of bedrooms are highly correlated with each other as shown in the correlation matrix. According to Burtler (1982), this problem could not be eliminated. Hoesli et al. (1997) decided to exclude the number of bedrooms from the hedonic regression model.
Hoesli et al. (1997) used the number of bathrooms as an indicator of age and comfort of the apartment. Furthermore, the existence of an elevator and the condition of the building were included. In their research also environmental variables were added. These variables measure the quality of the neighborhood and quality of the location within the neighborhood. The quality variable is measured by a dummy, which can take a value of poor, average and good.
The model explains almost 84% of total variance. The number of bathrooms turned out to be the most powerful explanatory variable is the regression. In line with existing hedonic literature, Hoesli et al. (1997) found that surface area has the highest level of significance and therefore the highest explanatory power. The researchers found that small surface apartments are relatively more expensive. This ‘overcharging’ is caused by high demand for these type of houses by students who receive housing allowances from the French government. Therefore, it is justifiable that Allen et al. (1995) included tenant characteristics in their model. Finally, a result of their research is that neighborhood quality is more important than the quality of the location in the particular neighborhood (Hoesli et al., 1997). The researchers conclude: ‘This model provides an accurate estimate for the rent of an apartment whose basic features are known’ (Hoesli et al., 1997).
In contrast to previous research, Wilson and Frew (2007) focus more heavily on distance variables, where the apartment rent gradient from urban theory stands central. This gradient explains the relationship between rents and the distance from
the apartment to the economic city center, which is based on the mono-centric model of Hoyt (1939). Rent gradient theory assumes an inverse relationship between rents and distance to the city center. Commuting costs to the city center play a crucial role in this model and increase in distance. Therefore, apartment rents decrease in commuting costs. This part will mainly focus on the relevant variables and interesting results gathered from their paper.
Wilson and Frew (2007) examine the change in the apartment-rent gradient of Portland from 1992 to 2002. Therefore, they performed a hedonic regression model for 1992 and 2002 including apartment characteristics, various distance variables and population growth to estimate these apartment rent gradients. Their dataset consists of 533 observations.
In addition to existing literature, these researchers analyzed both nominal rents and real rents. Furthermore, both a logarithmic- and a cubic function were used. This section will focus on the results from the logarithmic functional form. Table 1 only shows significant variables explaining nominal rents for one time period only, namely 2002.
The hedonic regression model of 1992 explains 66% of total variance, while the model of 2002 explains only 59% of total variance. The common used variable property age is not incorporated in their research. The results about the relation between distance and rents are in line with existing literature. A clear downward trend in rents could be observed as the distance to the city center or to an intersection of highways increases. Furthermore, the number of bathrooms has again high explanatory power. Additionally, the impact of bathrooms is stronger than the number of bedrooms, which serves as a proxy of apartment size. Finally, rents tend to increase in the existence of a laundry room and a swimming pool is not significant in explaining rents.
An interesting result from their paper is that apartment rents increased sharply in the city center over the 10-year period. This increase was mainly caused by population growth and intensive land restrictions. These restrictive urban growth boundaries are in line with the situation in the Netherlands. A chow-test also showed that the impact of distance variables became stronger over time and statistically
differed for the 1% level over the 10-year period. Therefore, they concluded that apartment rent gradients become steeper over time.
Finally the paper of Allen, Rutherford and Thomson (2009) is discussed. Allen et al. (2009) were the first researchers who provide an empirical model covering the relationship between asking rents, contract rents and time on the market. Their main aim was to address the rent setting mechanism. In house price literature, Kang and Gardner (1989) found a significant negative impact of time on the market on the actual selling price of a property. In addition, these researchers studied the effects across different price categories. The most powerful effect was related to properties with the lowest 25% of selling prices (Kang and Gardner, 1989).
Allen et al. (2009) performed an extensive hedonic regression model to measure the impact of; (1) rent and market variables, (2) property description variables and (3) listing and leased variables. This section focuses on the regression results where the log of contract rent is used as dependent variable. In addition, Allen et al. (2009) created three subgroups based on the level of their asking rents. The sample consists of single-family residential rental listings in North-Texas from 2003 and 2004. Their dataset is the largest till so far with approximately 20.000 listings.
In their research size is measured by square feet and divided by 100. Property age is divided by 10. In line with previous literature, age- and size of the residential rental unit are afresh powerful and significant at a 1% level in explaining rental listing rents (Allen et al., 2009).
Allen et al. (2009) also included several uncommon property specific variables such as: The existence of a security system and the permission of smoking. These variables both have a positive impact on contract rents. Subsequently, the impact of the number of bathroom is strong and significant.
In contrast to existing literature, the number of bedrooms is not significant for the sample as a whole. However it has a significantly positive effect for two subgroups with the lowest asking rents. The impact of the number of bedrooms is negative for the subgroup with the highest asking rents. Another interesting finding is that there exists a seasonality effect in their sample since rents are higher during the second- and third quarter of the year.
Finally, Allen et al. (2009) found that in case asking rents are lower than predicted asking rents, time on the market of the rental unit will decrease substantially (Allen et al., 2009). Time on the market is defined as the difference between the listing date and the lease contract date. The average number of days on the market was 72. More important, in case asking rents are set too high initially and therefore re-priced, the time on the market will increase. If the time on the market becomes larger, contract rents will decrease in the end (Allen et al. (2009).
At last, two papers that study regional differences between house prices are discussed shortly. Brounen and Huij (2004) studied regional differences between house prices in the Netherlands. They found that Dutch provinces react different on economic shocks. These economic changes could be related to GDP, interest rates, stock returns and unemployment rates. In addition, Capozza et al. (2002) studied this topic in the United States. In line with with Brounen and Huij (2004) he found that regional housing markets have a different sensitivity related to changes in economic fundamentals. Capozza et al. (2002) also included population growth as explanatory variable explaining regional variation in house prices.
3.4 Conclusion related literature !
This section will elaborate on variables of interest for this research based on existing related literature. The researchers listed above all performed a hedonic regression model. As illustrated in table 1 former rent level drivers do not provide a definitive set of variables to include in the hedonic regression model. However, based on the former seven articles a clear pattern could be observed. In case these explanatory variables are available for this thesis, they will be added in the final regression model.
As showed in table 1, the largest group of explanatory variables consists of house characteristics with fifth teen various variables. The most common house characteristics are property age, rental unit size, the existence of a fireplace and the number of bathrooms. Property age is significantly negative in most studies, while
size, a fireplace and the number of bathrooms are significantly positive in former studies. The existence of a swimming pool turned out to be the most common house feature variable in former research. As expected, this variable often has a significant positive impact on rents. All relevant house characteristics are included in this thesis as well. Furthermore, the number of stories is the most common building characteristic variable in discussed literature. However, Marks (1984) and Allen et al. (2009) found contrary signs.
An interesting result is that specific areas in the city have a significantly positive effect on rents. These findings of Marks (1984) and Sirmans et al. (1989) prove the importance of locational variables and the existence of market segmentation. The distance variable, which measures accessibility, is applied in the majority of the papers. This variable is often expressed as the distance to the highway, city center, bus line or university. Generally, these locational variables have a significant negative impact on rents. In addition, two researchers included a quarter dummy in their model to correct for seasonality effects. Table 1 shows that Q1 has a negative impact, while Q2 and Q3 have a positive impact on rents in their studies.
Subsequently, several additional variables are included in former research, such as: congestion, local owner or a security system. However, none of these variables showed up more than twice. This is probably because of poor data availability on these variables.
Finally, discussed related articles show the importance of submarkets in rental housing markets. Guntermann and Norrbin (1987), Hoesli et al. (1997) and Allen et al. (2009) examined different rent categories for instance. However, only Allen et al. (2009) found significant differences between rent level drivers for high-, medium- and low-rents rental units. These different rent segments have to be included in this thesis. In addition, Allen et al. (1995) found significantly different regression results with relation to various property types. He published that the hedonic pricing model has to be performed separately for every submarket to preclude for violating the linear regression model assumptions. According Hoesli et al. (1997) there exist substantial rent differences between different property types as well. In addition, Marks (1984), Sirmans et al. (1989) and Hoesli et al. (1997) showed that neighborhoods within a city
are an important explanatory variable to explain variation in rental unit rents. This thesis therefore includes neighborhood dummies to control for variation in rents. At last, Capozza et al. (2002) and Huij (2004) showed evidence that house prices could differ across regions. These regional differences are mainly caused by different sensitivities to changes in economic fundamentals. This thesis won’t elaborate on specific economic fundamentals explaining regional differences between rent levels. However, regional rent level differences are examined in detail.
4."Research(methodology!
The methodology section will further elaborate on the hedonic regression model that is applied in this thesis. Firstly, a conceptual model of the hedonic model is showed in figure 6. Secondly, the hedonic regression models are discussed. Thirdly, variables of interest are explained in further detail. Finally, this section focuses on several assumptions and considerations of the hedonic regression methodology.
Figure 6: Conceptual model
4.1 Hedonic regression model !
The primary focus of this thesis is to address the most important rent level drivers of rental units in the non-regulated rental housing sector of Amsterdam. Therefore, initial asking rents of 20.637 non-regulated rental units are analyzed, which were listed between Q2 2011 and Q2 2014. The model includes 41 explanatory variables (including TOM) to estimate the fair asking rents of non-regulated rented housing.
Section 5 further elaborates on the data. In addition, a comparison is made to test for differences in rent level drivers between two Dutch major cities, Amsterdam and Rotterdam. To correctly model this, a hedonic regression model is applied based on discussed related papers. The sample of Amsterdam consists of 13.152 observations, while the sample of Rotterdam consist of 7.485 observations.
The most appropriate functional form of the hedonic regression model is not clear in existing literature while results depend on model specification. Because most researchers have applied both the linear- and the semi-logarithmic functional form, this thesis will apply both functional forms. The model might benefit from the log form since the dependent variable rent is slightly skewed. An advantage of a semi-logarithmic function is that error terms are closer to normality (Francke, 2013). This is due to the fact that outliers have less impact. The hedonic regression model in this thesis regresses the dependent variable rent on various underlying rent level drivers as listed in table 2. The base regression model looks as follows:
!"#$! = !(!!", !!", !!", !!", !"#!, !!, !!")
Where rent stands for monthly rent per rental unit, S stands for a set of j specific house characteristics, F stands for a set of j house features, H stands for a set of j house types, L stands for a set of j locational variables, TOM stands for time on the market, T stands for time effectsand u stands for the error term.
The following four sections will discuss the application of this base regression model on the basis of the following four questions.
(1) What are the most important rent level drivers in the non-regulated housing sector in Amsterdam?
(2) Does the impact of rent level drivers differ between house types or rent segments in Amsterdam?
(3) Does the impact of rent level drivers differ between Amsterdam and Rotterdam?
(4)Does the impact of rent level drivers differ between house types or rent segments in Amsterdam and Rotterdam?
4.1.1.$Rent$level$drivers$in$Amsterdam$
To examine the specific impact of rent level drivers in the non-regulated rental housing sector in Amsterdam, the following base hedonic regression is performed. The extensive description of the variables is summarized in section 4.2.
(4.1.1)!!"#$%&!!"#"$!!"#$%&'(",!! = !0 + !! ! !!! !!,! + !! ! !!! !!,!+ !! ! !!! !!,!+ !! ! !!! !!,! + !! ! !!! !"#!,! + !! ! !!! !!,!+ !!
Where rent stands for monthly rent per rental unit, S stands for a set of j specific house characteristics, F stands for a set of j house features, H stands for a set of j house types, L stands for a set of j locational variables, TOM stands for time on the market, T stands for time effectsand u stands for the error term.
4.1.2.$Rent$levels$drivers$in$submarkets$in$Amsterdam$
Since rent level drivers could be different in submarkets, submarkets have to be examined separately. To examine these rent level drivers in several submarkets in the non-regulated rental housing sector in Amsterdam different hedonic regressions are performed. These submarkets consist of different house types and different rent categories, which is based on Allen et al. (1995) and Allen et al. (2009) respectively.
Only the two largest subgroups are included in this thesis, namely apartments and single-family houses. Rent categories consist of low-, medium- and high-segment with rents up to €1.000, between €1.000 and €1.600 and from €1.600. The base
regression model listed in 4.1.1. is applied to these submarkets in Amsterdam. Logically, the house type variable is excluded in case apartments and single-family houses are tested separately. In addition, a Chow-test is performed to determine whether submarkets for apartments and single-family houses within Amsterdam differ significantly. Section 4.1.3 explains the Chow-test in further detail.
4.1.3.$A$comparison$between$Amsterdam$&$Rotterdam$$
This section compares rent level drivers between both cities. Economic fundamentals are not incorporated in the model. To examine the differences of rent level drivers between Amsterdam and Rotterdam the following statistical tests are performed. Firstly, the base regression in section 4.1.1 is performed to examine rent level drivers for rental units in Rotterdam. Secondly, with the aid of a Chow-test differences between the models of Amsterdam and Rotterdam could be analyzed. Thirdly, a hedonic regression including interaction terms will be performed for the full sample of Amsterdam and Rotterdam to measure the differences between rent level drivers in Amsterdam and Rotterdam.
To examine the differences of rent level drivers between Amsterdam and Rotterdam or between house types and rent segments, a Chow-test is performed in this thesis. This statistical tool examines if there is sufficient evidence of a significant difference between two subsamples using an empirical model with similar regressors. Guntermann and Norrbin (1987) and Allen et al. (1995) also performed several Chow-tests to analyze structural breaks in the data. All variables of interest are included in the regression model and regressed on monthly rents. The intercept is included as well to test of both groups have an equal effect on rents. Both a linear and a semi-log functional form are employed. When Amsterdam and Rotterdam are compared, the zip codes and districts from both cities are excluded due to collinearity with the city dummy. To perform the Chow-test correctly, all explanatory variables are multiplied with a dummy of Amsterdam and also with a dummy of Rotterdam. Therefore, the difference in impact of these variables in both cities could be examined. The Chow test is F-distributed. The hypotheses are as follows:
H0: No significant differences between two subgroups
H1: Significant differences between two subgroups
The following multiple regression model includes interaction terms. To construct interaction terms, a binary variable (C) is multiplied with rent level drivers. The binary variable takes a value of 1 for the city of Amsterdam and 0 otherwise. Interaction terms measure the differences between the effects of rent level drivers in Amsterdam and Rotterdam for the full sample.
4.1.3! !!"#$%&!!"#"$!!"#$%&'("!!"##$%&'(,!! = !0 + !! ! !!! !!,! + !! ! !!! !!,! + !! ! !!! !!,!+ !! ! !!! !!,! + !! ! !!! !"#!,! + !! ! !!! !!,!+ !! ! !!! !!,!!!+ !! ! !!! !!,!!! + !! ! !!! !!,!!! + !! ! !!! !!,!!! + !! ! !!! !"#!,!!! + !! ! !!! !!,!!! + !!
!
4.1.4.$A$comparison$between$submarkets$in$Amsterdam$&$Rotterdam$The importance of submarkets in hedonic regression models is already addressed earlier. This section will therefore firstly analyze rent level drivers in different submarkets in Rotterdam with the aid of the model in section 4.1.1. Again, the house type variable is excluded in case apartments and single-family houses are tested separately. Subsequently, this section will compare rent level drivers between different submarkets in both Amsterdam and Rotterdam. The submarkets are again constructed based on house type and rent segment. In addition, a Chow-test is performed to examine if the data between two subgroups using equal hypotheses differs significantly. Finally, again a hedonic regression including interaction terms is performed. This regression model examines differences in rent level drivers between various submarkets in Amsterdam and Rotterdam.
4.2 Variables !
This section will shortly explain the variables of interest, listed in table 2. Firstly, the dependent variable in this thesis is the monthly asking rent of a non-regulated rental unit. This variable is expressed as a continuous variable and as a logarithmic function. Secondly, various house specific characteristics are included as independent variables into the regression model. The majority of them are based on existing related literature. All house specific variables are expected to have a positive impact on rents. However, the uncommon variable available for sharing is expected to have a negative impact on rents since the rental unit is intended for students. The inclusion of the furniture variable is based on Allen et al. (1995) and is divided into three classes; furnished, unfurnished and shell. Unfortunately, the powerful parameter property age is not incorporated in the hedonic regression model due to a lack of data. Finally, the variable that measures size of the rental unit is transformed to a step function divided in four categories. The step function is performed to allow non-linearity (Stock and Watson, 2012). In the results section either the continuous form or the step function is chosen to explain variation in rents. According to existing literature, all house features are expected to have a positive impact on rent levels.
Since there exist substantial rent differences across house types according to Allen et al. (1995), a distinction is made between single-family houses and apartments. In addition, locational dummies are included in the hedonic regression model since this precludes for an incomplete specification in which neighborhood information is omitted. The locational variable implicitly controls for distance to the center, quality differences between neighborhoods or the average income within the neighborhood. Both Marks (1984) and Hoesli (1997) showed the significant impact of specific areas or neighborhoods in the city on rents. Therefore, this thesis will include zip code dummies and district dummies to control for differences between neighborhoods.
Subsequently, the listing period of the rental unit on the market is included in the model. Allen et al. (2009) defined this variable as the difference between the listing date and the lease contract date. In contrast to Allen et al. (2009), this thesis will examine the impact of time on the market on asking rents. A positive relationship
is assumed since the longer the time on the market, the higher the initial asking rent. However, the causal relationship of days on the market on initial asking rents is incorrect since initial asking rents affect the total days on the markets. In addition, the variable time on the market (TOM) could be imprecise since rented rental units are not always deregistered from Pararius immediately.
Finally, two types of time dummies are included in the model. On the one side a time period dummy is added. Based on Francke (2013) this time effect serves as a proxy for market conditions during the sample period. Based on Marks (1984), a quarterly dummy variable is added to control for seasonality effects during a particular year on the other side. Unfortunately no accessibility variables are included caused by a lack of data.
Table 2: Variable definition Variables: Rents Rents per m2 Rents <1000 Rents 1000-1600 Rents >1600 House specific characteristics: Size Size < 60 Size 60-80 Size 80-100 Size 100-120 Size < 120 #Bedrooms 1 Bedroom 2 Bedrooms 3 Bedrooms 4 Bedrooms >4 Bedrooms #Bathrooms 1 Bathroom >1 Bathrooms Garden Roof Terrace Balcony Bath Separate Shower Fireplace Air condition Furnished Unfurnished Definition:
Asking rents (monthly) provided by Pararius Asking rents per m2 (monthly) provided by Pararius Asking rents (monthly) less than €1.000
Asking rents (monthly) between €1.000 and €1.600 Asking rents (monthly) more than €1.600
Rental unit size expressed in square meters
Dummy variable equal to 1 if size is less than 60 m2, 0 otherwise
Dummy variable equal to 1 if size exceeds 59m2 and is less than 80 m2, 0 otherwise Dummy variable equal to 1 if size exceeds 79m2 and is less than 100 m2, 0 otherwise Dummy variable equal to 1 if size exceeds 99m2 and is less than 120 m2, 0 otherwise Dummy variable equal to 1 if size exceeds 119 m2, 0 otherwise
Number of bedrooms in rental unit
Dummy variable equal to 1 if rental unit has 1 bedroom, 0 otherwise Dummy variable equal to 1 if rental unit has 2 bedrooms, 0 otherwise Dummy variable equal to 1 if rental unit has 3 bedrooms, 0 otherwise Dummy variable equal to 1 if rental unit has 4 bedrooms, 0 otherwise
Dummy variable equal to 1 if rental unit has more than 4 bedrooms, 0 otherwise Number of bathrooms in rental unit
Dummy variable equal to 1 if rental unit has 1 bathroom, 0 otherwise
Dummy variable equal to 1 if rental unit has more than 1 bathroom, 0 otherwise Dummy variable equal to 1 if rental unit has a private garden, 0 otherwise Dummy variable equal to 1 if rental unit has a private roof terrace, 0 otherwise Dummy variable equal to 1 if rental unit has a private balcony, 0 otherwise Dummy variable equal to 1 if rental unit has private bath, 0 otherwise
Dummy variable equal to 1 if rental unit has separate shower instead of a shower in bath, 0 otherwise
Dummy variable equal to 1 if rental unit has a fireplace, 0 otherwise Dummy variable equal to 1 if rental unit has an air-condition, 0 otherwise Dummy variable equal to 1 if rental unit is furnished, 0 otherwise
Shell Available For Sharing House Features: Parking Place Garage Elevator Storage Pool House type variable: Apartment House Locational variables: Zip code District City Time on the market: TOM Time effects: Quarter_year Q1 Q2 Q3 Q4
Dummy variable equal to 1 if rental unit has unfinished interior, 0 otherwise
Dummy variable equal to1 if it is allowed to share the rental unit with another tenant, 0 otherwise
Dummy variable equal to 1 if rental unit has private parking place, 0 otherwise Dummy variable equal to 1 if rental unit has a garage, 0 otherwise
Dummy variable equal to 1 if building has a elevator, 0 otherwise
Dummy variable equal to 1 if rental unit has a private storage, 0 otherwise Dummy variable equal to 1 if rental unit has a pool, 0 otherwise
Dummy variable equal to 1 if rental unit is an apartment, 0 otherwise Dummy variable equal to 1 if rental unit is a house, 0 otherwise
Dummy variable equal to 1 if rental unit is located in particular Zip code, 0 otherwise Dummy variable equal to 1 if rental unit is located in a particular district, 0 otherwise Dummy variable equal to 1 if rental unit is located in Amsterdam, 0 for Rotterdam
Number of days a rental unit was listed on Pararius. (Time On Market)
Dummy variable equal to 1 if rental unit was listed in this quarter for a particular year, 0 otherwise (period from Q2-2011 to Q2-2014)
Dummy variable equal to 1 if rental unit was listed in the 1st quarter, 0 otherwise Dummy variable equal to 1 if rental unit was listed in the 2nd quarter, 0 otherwise Dummy variable equal to 1 if rental unit was listed in the 3rd quarter, 0 otherwise Dummy variable equal to 1 if rental unit was listed in the 4th quarter, 0 otherwise
